Capability-Aware Heterogeneous Control Barrier Functions for Decentralized Multi-Robot Safe Navigation

A team of researchers led by Joonkyung Kim has published a new paper introducing the Capability-Aware Heterogeneous Control Barrier Function (CA-HCBF) framework, a decentralized system designed to solve a critical problem in multi-robot navigation. Current methods often assume all robots are identical, which causes issues when coordinating teams with different physical capabilities—like drones, wheeled robots, and legged robots moving together. When safety constraints aren't tailored to each robot's unique dynamics, some agents receive avoidance commands they physically cannot execute, leading to collisions or system-wide deadlocks.

The CA-HCBF framework addresses this by first creating a unified mathematical representation that works for both holonomic (can move in any direction) and nonholonomic (like cars) robots. More importantly, it introduces a novel "directional capability metric" that quantifies each robot's ability to follow its intended path. This metric enables a pairwise responsibility allocation system that distributes the "safety burden" of collision avoidance proportionally to what each robot can actually achieve. A feasibility-aware clipping mechanism further ensures no robot is asked to perform maneuvers outside its physical limits.

Simulations with up to 30 heterogeneous robots and a physical multi-robot demonstration showed the framework outperforms existing baselines. It successfully maintained safety (zero collisions) while also improving overall task efficiency, as robots spent less time stuck in avoidance maneuvers or deadlocks. The work, available on arXiv, validates a practical approach for real-world applications where robots with distinct kinematics must cooperate in crowded spaces, from warehouse logistics to search-and-rescue missions.